Public SaaS enterprise software companies saw their stock prices plummet. This is because of claims that SaaS suppliers will be disrupted by Generative AI. Wall Street believed that Anthropic, OpenAI, and other generative AI solutions’ ability to rapidly generate software code will lead companies to build their own applications, thereby avoiding costly SaaS fees. Enterprise companies’ stock prices are drifting higher but have not fully recovered.

More recently, the AI goalpost has been moved. Now, according to UBS, it is Agentic AI that matters. UBS is a major Swiss multinational investment bank and financial services company. One of their services is stock investment advice. UBS downgraded SAP from Buy to Neutral on August 26, 2026, citing concerns that the company’s agentic AI rollout is too slow. SAP is the world’s largest enterprise software company.

I have believed that Wall Street does not fully understand the supply chain software business, particularly supply chain planning. In this, I agree with Anaplan’s CEO Charlie Gottdiener. Anaplan is a large player in the enterprise market, one of the relatively few enterprise software companies that generate over a billion dollars annually. Anaplan offers a cross-functional scenario planning and analysis platform whose applications span finance, supply chain, human resources, and sales and marketing. They have had particularly strong traction in supply chain in the last few years.

Anaplan’s CEO argued , “Companies that have certain characteristics in the SaaS space are going to do well.” Anaplan’s cross-functional scenario-planning and analysis platform performs deterministic calculations. Generative AI, AI based on large language models, is poor at this. LLM AI provides probability-based answers. LLM AI can be used to improve its solution set. But it will not replace them.

The answers provided by the Anaplan engine must be 100% correct. Large Language Model AI can’t provide this level of accuracy. In the supply chain realm, for example, a company that ships inventory to the wrong distribution center 5% of the time is not doing well. Gottdiener exclaimed, “That would be a terrible outcome!” Similarly, a financial report that looks great but is riddled with errors won’t cut it.

When it comes to deterministic calculations, a supply chain planning engine that uses optimization would, I thought, be particularly hard for genAI to replicate. From an agentic perspective, the optimization engine would be a very large agent that would need to be part of the agentic orchestration around planning. It would not be a typical agent, and it would be impossible for Anthropic or OpenAI to produce, I thought.

Perhaps, Wall Street understands this argument. Kinaxis Inc. (TSX: KXS) has not seen their stock price go down significantly year-to-date. In 2026, the stock is up about 1.87%. Kinaxis, a leading supply chain planning vendor,

David Simchi-Levi, a distinguished professor at Purdue University, co-authored an academic paper arguing that it is now technologically feasible to transition from optimization to AI-driven decision-making in supply chain planning.

Mr. Simchi-Levi has a distinguished career in supply chain planning. He was formerly the chief scientist at ILOG. In the early days of supply chain planning, all the planning vendors used ILOG to build their optimization engines. He went on to cofound a supply chain company while he was a professor at MIT. That company was acquired by the global consulting giant Accenture.

Unless you have a doctorate in operations research, and I do not, it would be difficult to make sense of this article. Fortunately, Simchi-Levi posted a summary on LinkedIn.

He points out that for decades, optimization engines have been used for complex supply chain decisions. Often, SCP optimization involves a solution that lowers costs while achieving a specified customer service level. Supply planning optimization generally does not yield a single best solution to a complex planning problem.

Because of the tens of thousands of variables and constraints, a supply chain problem can mathematically explode. Coming up with the one best solution could take a planning engine 10,000 years to produce. So, what optimization in supply planning often did was converge on a very, very good answer in a certain amount of time, often a few hours.

The professor rightly points out that traditional optimization can be too slow for real-time applications. “In this work,” Simchi-Levi wrote, “we ask a simple question: Can AI learn to make high-quality operational decisions at a scale and speed that traditional optimization struggles to achieve?” They used what they are calling an “OR-Transformer” to produce supply chain answers. This is “a deep reinforcement learning framework that combines a Transformer architecture with the structure of the underlying inventory dynamics.”

Optimization engines often use a technique called mixed integer linear programming. In experiments with over 1,000 inventory items, their approach, he claims, “increasingly outperforms both learning-based methods and rolling-horizon MILP approaches as the problem scales.” Now, I should mention that when it comes to optimizing inventory levels, 1,000 items is not very many. Many large enterprises have tens of thousands of stock-keeping units. Simchi-Levi says their solution could make decisions in less than half a second, compared with nearly 49 hours for powerful MILP solvers, while achieving a 19.1% lower inventory cost.

Leading SCP vendors would not take 49 hours to solve this kind of problem, but it could take an hour or more. As to lower inventory costs, this is a comparison of their solution to their MILP solver. They are not benchmarking against off-the-shelf supply chain planning solutions.

Nevertheless, while not an apples-to-apples comparison, SCP vendors should be worried. The moat surrounding supply planning engines appears to be disappearing.